Privacy Policy
Last updated:
This Privacy Policy describes how Mergestorm ("we," "us," or "our") may collect, use, and share information when you use our website and application.
Information we may collect
Mergestorm is an AI code-review product for GitHub pull requests. Depending on how you use the service, we may process: account details you provide (such as name and email from GitHub, Google or Sign in with Apple), GitHub repository metadata and pull request content (diffs, commits, linked issues) needed to run reviews and generate inline comments and check runs, billing and subscription data via our payment processor, usage and credit balances, and technical logs (for example IP address, device type, and approximate location) to operate and improve the service.
When you create an account, we record how you reached us so we can tell which links bring sign-ups: campaign tags in the link you followed (such as utm_source or ref), the domain of the site that sent you, and the page you signed up from. If you have not accepted analytics cookies, we read these from the page at the moment you click sign in and keep them on our servers only, without cookies or browser storage.
If you use the Mergestorm Mac or iPhone app, we record each day you use it, each chat turn you send from it (which chat, and whether it completed or failed, but not what you wrote), and the app version. For each device you sign in on, we keep the device name, whether it is a Mac or an iPhone, the app version, and when it signed in and last refreshed its session.
When you download an app from our website, we record which app and file, where on the page you clicked, the site that sent you, your country, your device type and browser, a one-way visitor hash like the one described under Cookies and consent, and your account if you are signed in.
Support and quality review
When you chat with Breeze, each message you send and Breeze's reply, including a reply that failed partway, is posted to our team's internal Slack workspace so we can support you and check the quality of answers. The post includes your email, the chat's ID, the app and version you used, and the number of files you attached. When a message has files attached, we post your message and the number of files, but not the files or the reply.
Feedback you send from the Mac or iPhone app is posted to the same Slack workspace with its text, your email and GitHub username, the app and version, and the conversation it came from.
Mergestorm administrators can open users' chats and the files uploaded to them, for support, abuse prevention and quality work.
AI model providers
To review and fix your pull requests, we send repository content to OpenAI, DeepSeek and Anthropic, or to other providers when a configured model or runner override is set. This includes diffs, file contents, commit messages, and pull request and comment text. ElevenLabs receives the audio you dictate, a fixed list of Mergestorm vocabulary terms, and, for some requests, the language you dictate in. We use each provider's business API.
- OpenAI (GPT-6 Luna, GPT-5.6 Luna): Vortex follow-up reviews of later pushes, the second model on a pull request's first Vortex review when swarm mode is Normal or Max, Vortex's pre-review briefing, repository overview and finding consolidation, Vortex comment triage, and the Vortex summary of your repository's rules files. GPT-6 Luna answers Cyclone comment triage when Claude Haiku 5.5 does not, and GPT-5.6 Luna is the backup when Claude Haiku 5.5 does not answer a Vortex review. In dashboard chat, GPT-5.6 Luna sorts each message, explains findings, summarizes long threads, and answers the signed-out demo chat.
- DeepSeek (DeepSeek V4 Flash, DeepSeek V4 Pro): dashboard chat answers, the title of a review unit's land PR, and the third model on a pull request's first Vortex review when swarm mode is Max. Requests go to DeepSeek's own API. DeepSeek is based in China.
- Anthropic (Claude Haiku 5.5): a pull request's first Vortex review uses Claude Haiku 5.5, and so do Vortex issue reviews, the extra passes and provider retries of any Vortex review, and a follow-up review when GPT-6 Luna does not answer. Claude Haiku 5.5 is Cyclone's default for patch runs, comment triage, and repository-rules summaries. Configured model or runner overrides may send Cyclone content to other providers.
- ElevenLabs (Scribe v2 and Scribe v2 Realtime): when you dictate a chat message on the website or in the Mac or iPhone app, the audio you record is sent through our servers to ElevenLabs to be turned into text. We do not store the audio or log the text. The text goes into the message box and is sent only when you send the message.
Surge CI runners, the merge queue, and Auto land do not send your code to an AI model provider.
Model training
Mergestorm does not train models on your code unless Share Data is on for your account. We send code to OpenAI, DeepSeek and Anthropic through their business and API offerings, and to other providers when a configured model or runner override is set.
We ask about Share Data right after sign-up. The switch on that screen starts on, except when you sign up from the EU, the EEA, the United Kingdom or Switzerland, or when we cannot tell your country, where it starts off. Nothing is shared until you press Continue there. While Share Data is on, we may use your code, chats, and activity in Mergestorm to improve the product and the AI behind it. You can turn it off at any time in Account.
How long we keep data
- Your account, settings, reviews, findings, chats, and history: kept until you delete your account.
- The diff, files, and context you send to the review API: deleted 7 days after the review finishes. The review itself and its findings stay.
- Stored copies of the GitHub events behind each review or patch job: deleted 90 days after the job finishes. Events we need to run a review of the same pull request again are kept past that only while we still track the pull request as open (not a draft and not closed), or while it is part of a stack you have not archived.
- Vortex's working notes for a pull request, which can quote your code: deleted 30 days after the pull request was last reviewed. Notes are kept past that only while we still track the pull request as open, or while it is part of a stack you have not archived.
- Copies of your code on our review workers (snapshots and work folders): deleted 7 days after they were last written.
- Cyclone checkouts of your repository: deleted 7 days after Cyclone last used them. The temporary home folder for each patch run is deleted when the run ends.
- Model responses, which can quote your code, logged with the repository and pull request they belong to: deleted about 14 days after they are written.
- Service logs: rotated by size. We keep the current file and five 5 MB archives per service, so older lines are deleted as new ones arrive. There is no fixed time limit.
- Surge runner machines: the job folder is wiped after every job. Each machine is deleted when there is no work for it. A machine older than 8 hours is replaced when a new one can start. Until then it can keep taking jobs, and it is deleted once it is idle, so there is no fixed maximum lifetime. If our controller stops responding, a machine older than 8 hours is deleted at once and the jobs on it are stopped and run again.
How this data is protected, and who can read it, is on the Security page.
Third-party services
We may rely on vendors for hosting, authentication, payments, analytics, email delivery, or error reporting. For example, we may use Google Analytics to better understand aggregate usage and interaction patterns (such as clicks and page behavior) so we can improve product usability. We use Google Ads and X (Twitter) ads to measure which ads bring sign-ups, as described under Cookies and consent, and Slack for the internal support and quality review described above. Those providers process data according to their own policies and our agreements with them, only as needed to deliver the product.
Marketing email
We send marketing email, such as product news, offers, and free credits, only if you opt in. The box to opt in at sign-in is never pre-ticked. We record when you opted in, where (sign-in or Account settings), and the wording you agreed to.
You can withdraw consent at any time with the Email toggle in Account settings or the unsubscribe link in any of our emails. Withdrawing is free and takes effect for the next email we send.
Transactional and account email, such as the welcome message and subscription and billing notices, is separate. We send it to run your account whether or not you opt in to marketing email.
Account deletion
You can request irreversible account deletion from Account settings. We revoke sessions and product API keys, stop your active work, cancel subscriptions, and delete your account-owned database records, including chats, review jobs, history, settings, and session metadata. Failed steps retry automatically. You can also delete your account from the Mergestorm Mac and iPhone apps.
If you sign in with Apple, Apple shares your name and email with us, or a private relay address if you choose to hide your email. We keep the Apple sign-in token encrypted so we can revoke it with Apple when you delete your account; deletion revokes it before your data is removed.
Shared repository records and other users’ installation, repository, and history records remain. Worker snapshots, checkouts, prompts, and agent homes are queued for separate cleanup; accepting an account deletion request does not mean those files are gone. This flow does not delete data already held by GitHub or other vendors.
Payment providers retain invoices and billing records required for accounting and legal obligations. We retain email suppression records to honor opt-outs and minimal deletion tracking identifiers, including worker cleanup identifiers until cleanup is acknowledged.
We keep a security audit record of administrator access, GitHub installation transfers, session and API key revocations, and account deletion requests for 400 days, including after account deletion. It holds account and object identifiers, not tokens, prompts, or code.
Contact
For privacy-related questions, contact us at contact@mergestorm.ai.
See also our Terms of Service.